CN107333282B

CN107333282B - 5G terminal universal platform optimization method and system based on GPP

Info

Publication number: CN107333282B
Application number: CN201710415240.1A
Authority: CN
Inventors: 唐彦波
Original assignee: Huizhou TCL Mobile Communication Co Ltd
Current assignee: Huizhou TCL Mobile Communication Co Ltd
Priority date: 2017-06-05
Filing date: 2017-06-05
Publication date: 2021-02-19
Anticipated expiration: 2037-06-05
Also published as: EP3637830A4; EP3637830A1; CN107333282A; WO2018223932A1; US20200183741A1

Abstract

The invention discloses a 5G terminal universal platform optimization method and a system based on GPP, wherein different priorities are allocated to programs of different modules in a base station, and the highest priorities are allocated to control channels and related programs of a control processing flow; after the task with the high priority is processed, the task with the low priority is processed; segmenting the tasks according to the attribute of each task, and distributing the subtasks to different threads; allocating a time budget to each subtask, marking each subtask with a time stamp in the processing flow, and comparing the time stamp of each subtask with the allocated time budget to decide to continue execution or terminate in advance; the real-time processing is high, the feedback time delay is low, the strict requirements of high real-time performance and low delay in mobile communication are met, and the conventional general platform is greatly optimized.

Description

5G terminal universal platform optimization method and system based on GPP

Technical Field

The invention relates to the technical field of 5G, in particular to a 5G terminal universal platform optimization method and system based on GPP.

Background

Currently, the global technical research on 5G is actively being carried out, but 3GPP standardization is also being carried out synchronously, and there is no standardized version so far. Most of the manufacturers engaged in 5G research agree that 5G can gradually enter the commercial stage and enter people's lives worldwide by 2020. Based on the uncertainty of the 5G protocol, the software architecture design of the test terminal is challenged. Different from the traditional system based on FPGA, special chip or DSP, the open type 5G wireless system based on pure software architecture realized by a general processor can conveniently use various mature software engineering methods, thereby improving the software development efficiency and quality; however, the open architecture based on open pure software also faces many problems in software implementation, such as real-time processing of LTE and 5G protocol stacks, HARQ feedback delay, implementation of multi-terminal simulation, and the like, which brings great inconvenience.

Thus, the prior art has yet to be improved and enhanced.

Disclosure of Invention

In view of the defects of the prior art, the invention aims to provide a 5G terminal general platform optimization method and system based on GPP, and aims to solve the problems of low real-time performance and high time delay of the existing 5G terminal general platform based on a general processor.

In order to achieve the purpose, the invention adopts the following technical scheme:

A5G terminal universal platform optimization method based on GPP includes:

A. distributing different priorities to programs of different modules in the base station, and distributing the highest priorities to control channels and related programs of a control processing flow; after the task with the high priority is processed, the task with the low priority is processed;

B. segmenting the tasks according to the attribute of each task, and distributing the subtasks to different threads;

C. assigning a time budget to each subtask, time stamping each subtask in the processing flow, and comparing the time stamp of each subtask with the assigned time budget to decide to continue execution or terminate prematurely.

The 5G terminal general platform optimization method based on the GPP is characterized in that the 5G terminal general platform based on the GPP simultaneously supports static scheduling and dynamic scheduling.

The 5G terminal universal platform optimization method based on GPP, wherein the step B further comprises:

b1, allocating the priorities of the tasks and the threads.

b2, pre-processing tasks in the background thread.

The 5G terminal universal platform optimization method based on GPP, wherein the step C further comprises:

c1, monitoring the task execution by the task controller, and communicating with the scheduler to increase or decrease the tasks processed by the physical layer.

The 5G terminal universal platform optimization method based on GPP comprises the steps that an FPGA acceleration unit is adopted to build a heterogeneous computing platform, and the FPGA is used for accelerating the baseband signals so as to reduce the computing burden of a universal processor; directly accessing a memory of the general server platform by adopting a DMA (direct memory access) technology through a PCI-E (peripheral component interconnect-express) interface to read and write data so as to realize high-speed data interaction between the general processor and the acceleration unit; and parallel processing of single-instruction multi-path data streams is completed by adopting SIMD (single instruction multiple data) instructions supported by a general processor, wherein the instruction set-based software acceleration method comprises bit-level acceleration, symbol-level acceleration and/or sampling-level acceleration.

A5G terminal universal platform optimization system based on GPP comprises:

the priority module is used for distributing different priorities to programs of different modules in the base station and distributing the highest priorities to the control channels and the related programs of the control processing flow; after the task with the high priority is processed, the task with the low priority is processed;

the task segmentation and distribution module is used for segmenting the tasks according to the attribute of each task and distributing the subtasks to different threads;

and the task execution module is used for allocating a time budget to each subtask, marking a time stamp for each subtask in the processing flow, and comparing the time stamp of each subtask with the allocated time budget to decide to continue execution or terminate in advance.

The 5G terminal universal platform optimization system based on GPP further comprises:

and the preprocessing module is used for preprocessing tasks in background threads.

and the monitoring module is used for monitoring the execution condition of the tasks through the task controller and communicating with the scheduler to increase or decrease the tasks processed by the physical layer.

The 5G terminal universal platform optimization system based on GPP comprises an FPGA acceleration unit, a baseband signal processing unit and a general processor, wherein the FPGA acceleration unit is adopted to build a heterogeneous computing platform, and the FPGA is used for accelerating the processing of the baseband signal so as to reduce the computing burden of the general processor; directly accessing a memory of the general server platform by adopting a DMA (direct memory access) technology through a PCI-E (peripheral component interconnect-express) interface to read and write data so as to realize high-speed data interaction between the general processor and the acceleration unit; and parallel processing of single-instruction multi-path data streams is completed by adopting SIMD (single instruction multiple data) instructions supported by a general processor, wherein the instruction set-based software acceleration method comprises bit-level acceleration, symbol-level acceleration and/or sampling-level acceleration.

Compared with the prior art, the method and the system for optimizing the 5G terminal universal platform based on the GPP, which are provided by the invention, have the advantages of high real-time processing and low HARQ feedback delay, meet the strict requirements of high real-time performance and low delay in mobile communication, facilitate the realization of multi-terminal simulation, greatly optimize the existing universal platform and bring great convenience.

Drawings

Fig. 1 is a flowchart of a method for optimizing a 5G terminal universal platform based on GPP provided in the present invention.

Fig. 2 is a structural block diagram of a GPP-based 5G terminal universal platform optimization system provided in the present invention.

Fig. 3 is a schematic diagram of an architecture of a 5G terminal universal platform optimization system based on GPP provided in the present invention for simulating a large number of terminals.

Detailed Description

The invention provides a method and a system for optimizing a 5G terminal general platform based on GPP. In order to make the objects, technical solutions and effects of the present invention clearer and clearer, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

Referring to fig. 1, the present invention provides a 5G terminal universal platform optimization method based on GPP, where the 5G terminal universal platform optimization method based on GPP includes:

s100, distributing different priorities to programs of different modules in a base station, and distributing the highest priorities to control channels and related programs of a control processing flow; after the task with the high priority is processed, the task with the low priority is processed;

s200, segmenting the tasks according to the attribute of each task, and distributing the subtasks to different threads;

s300, allocating a time budget to each subtask, marking a time stamp for each subtask in the processing flow, and comparing the time stamp of each subtask with the allocated time budget to decide to continue execution or terminate in advance.

The above steps are described in detail with reference to specific examples.

In step S100, different priorities are assigned to programs of different modules in the base station, and the highest priority is assigned to a control channel and a related program of a control processing flow; and processing the task with low priority after the task with high priority is processed. Specifically, the present invention is a software optimization method based on a gpp (general Purpose processor) general processor platform, and due to the strict requirements of high real-time performance and low delay in mobile communication, the software architecture of a base station based on an Intel general processor architecture is different from the conventional software architecture. For example, in the LTE standard, each subframe occupies l milliseconds, which requires that the software program at the base station must complete decoding of the uplink channel and responding to ACK/NACK (acknowledge/non-acknowledge) transmission by the UE (terminal) within 3 milliseconds. The time occupied by each subframe of the 5G NR is much less than 1 millisecond, and the feedback time of HARQ (Hybrid Automatic Repeat reQuest, a technology formed by combining forward error correction coding FEC and Automatic Repeat reQuest ARQ) is more strict than that of LTE.

In addition, the base station based on the Intel general processor architecture also differs from the conventional base station based on the digital signal processor in software architecture. In a DSP-based software radio implementation, the delay of the digital signal processing is almost fixed, so the flow of the program processing can be tightly controlled. In the terminal simulator based on the Intel architecture, due to the multi-core and multi-thread architecture of the operating system, a program may jitter when processing the same functional module. One solution is to bind the specified task to a core, and let this core only run fixed programs.

How to fully utilize the multiple cores in the processor is a challenge in the design of base station software architecture based on the intel general purpose processor architecture. The software architecture must be scalable to support different processors and any number of cores. For example, after the terminal program configuration is completed, it may process one sector of traffic on a dual-core processor, or may process three sectors of traffic on a quad-core processor.

The design principle of the base station software architecture based on the Intel general processor architecture is that the processing capacity of a multi-core processor is utilized to the maximum extent on the basis of meeting the real-time requirement in a protocol. To this end, different priorities are assigned to the programs of different modules in the base station. To ensure proper operation of the communication protocol, the control channel and associated procedures controlling the process flow must have the highest priority. On this basis, the best effort is made to process the correlation of the data channel, depending on the capabilities of the processor. This means that both a dual-core processor and a quad-core processor can support the processing of the 20MHz TDD LTE protocol, but the quad-core processor can provide higher data throughput. Preferably, for optimization, the GPP-based 5G terminal generic platform supports both static scheduling and dynamic scheduling.

In step S200, the task is divided according to the attribute of each task, and the subtasks are allocated to different threads. Specifically, for the purpose of optimization, the GPP-based 5G terminal universal platform of the present invention can rapidly and efficiently segment tasks according to the attribute of each task and flexibly allocate the tasks to different threads when executing the tasks.

Preferably, the step S200 further includes: s201, distributing priorities of tasks and threads. In particular with regard to priority assignment of tasks and threads. As described above, the control channel handler needs to assign a higher priority. However, some tasks do not have strict real-time requirements, so that the tasks can be assigned with lower priorities, and the tasks with lower priorities are processed after the tasks with higher priorities are processed. For example, the response in the PRACH (Physical Random Access Channel) Channel can be processed for a longer period of time.

In step S300, a time budget is allocated to each subtask, each subtask is time-stamped in the processing flow, and the time stamp of each subtask is compared with the allocated time budget to decide to continue execution or terminate prematurely. Specifically, the task is divided into a plurality of subtasks and allocated in step S200, then a time budget is allocated to each subtask and the processing time of each subtask is recorded in step S300, and if the time budget is exceeded, the execution is continued, and if the time budget is not exceeded, the execution is terminated in advance. In actual application, the task manager needs to allocate a time budget to each subtask. The software will timestamp the subtasks in the process flow and compare them with the allocated time budget to decide whether to continue execution or terminate prematurely. The software will first process the high priority tasks and then process the other tasks with the best effort possible. For example, in the Physical layer of the LTE terminal receiving end, IFFT (inverse fast fourier transform) operation is performed first, then a pair of Control Channel PDCCH (Physical Downlink Control Channel) is decoded one by one, and finally a data Channel PDSCH (Physical Downlink Shared Channel) is processed according to the remaining time.

Preferably, the step S200 further includes:

s202, preprocessing tasks in background threads. In particular, it is the task that can be pre-processed in the background thread. Many tasks of the transmitting end can be calculated in advance, for example, the reference signals of the next 20 frames can be calculated in advance and stored in the memory. In addition, there is no very strict real-time requirement for transmission of new data in the PDSCH channel, so that the scheduler can decide when to transmit after the modulation and coding implementation is calculated. This pre-processing mechanism can reduce the occurrence of jitter and provide more room for high priority or time limited tasks.

Preferably, the step S300 further includes:

s301, the task controller monitors the execution condition of the tasks and communicates with the scheduler to increase or decrease the tasks processed by the physical layer. Specifically, the task controller monitors the execution of the tasks and communicates with the scheduler to increase or decrease the number of tasks processed by the physical layer. For example, when there are a large number of physical layer transmission tasks that cannot be completed within the allocated time budget, the scheduler may allocate fewer tasks to future upstream processing. Thus, the throughput that can be supported on a particular platform will depend on the processing power of the processor. In addition, the task controller must also balance the load between the different cores. There is also a need to provide a mechanism to prevent jitter from occurring during scheduling. This is the role of the intelligent scheduler and task controller in the general processor platform of the present invention.

In practical application, the physical layer, the MAC layer and the RLC layer of the 5G terminal universal platform based on the GPP are designed in a cross-layer mode. These three layers are highly coupled in the LTE protocol, and although functionally independent, are coupled together from the task execution perspective. In the conventional implementation mode, different layers use independent hardware structures, and the movement of data among the layers causes redundant delay and waste. A terminal based on the Intel general processor architecture can conveniently implement a cross-layer design because the entire protocol stack runs on one processor.

Preferably, for optimization, in terms of hardware, the GPP-based 5G terminal universal platform adopts an FPGA acceleration unit to build a heterogeneous computing platform, and performs acceleration processing on a baseband signal through an FPGA to reduce the computing burden of a universal processor; directly accessing a memory of the general server platform by adopting a DMA (direct memory access) technology through a PCI-E (peripheral component interconnect-express) interface to read and write data so as to realize high-speed data interaction between the general processor and the acceleration unit; and parallel processing of single-instruction multi-path data streams is completed by adopting SIMD (single instruction multiple data) instructions supported by a general processor, wherein the instruction set-based software acceleration method comprises bit-level acceleration, symbol-level acceleration and/or sampling-level acceleration.

Specifically, in order to achieve the technical effects corresponding to the above flow steps, on one hand, an FPGA acceleration unit is adopted to build a heterogeneous computing platform, and the FPGA is used for accelerating certain baseband signals which are relatively simple in computation but large in computation, so that the computation burden of a general processor is reduced; designing a PCI-E interface, and directly accessing read-write data to the memory of the general server platform by adopting a DMA (direct memory access) technology to realize high-speed data interaction between the general processor and the acceleration unit; the forward link radio frequency and the general server baseband processing interface are connected by adopting a mature CPRI interface; and providing a hardware architecture of the whole open 5G universal platform terminal simulator, and realizing the software-defined physical layer, especially the baseband processing function. On the other hand, the parallel processing of the single-instruction multi-path data stream is completed by adopting SIMD instructions (MMX, SSE, SSE2, SSE3, SSE4, AVX, AVX2 and the like) supported by a general processor in combination with the architectural characteristics of the Intel processor. The software acceleration method based on the instruction set comprises the following steps: 1) bit-level acceleration: Look-UP Tables (Bit level-Look UP Tables, LUT); the LUT is a compromise operation performed after considering computational complexity and spatial complexity, and the LUT operation can greatly reduce the on-line processing delay by replacing the conventional bit operation. And bit-level operations such as CRC check and de-check, scrambling and descrambling, rate matching and de-matching can be accelerated by adopting an LUT mode. 2) Symbol-level acceleration: single Instruction Multiple Data (SIMD) instructions, wherein the Intel CPU has a special SIMD (Single instruction multiple data) instruction set to accelerate signal processing for Symbol-level operations. SIMD performs the same operation mainly for symbol-level data repetition. One instruction of SIMD can process several operations, the operation cost (calculation resource) is small, and the bit bandwidth is fully used, so that it can raise CPU efficiency obviously. And symbol-level operations such as modulation and demodulation, precoding, MIMO, channel estimation and the like can adopt a SIMD mode. 3) Sampling stage acceleration: an Intel comprehensive Performance primitive (IPP), wherein the Intel developed comprehensive Performance primitive IPP is a set of software function library spanning platforms and operating systems, and can implement operations such as signal processing, image processing, multimedia, vector processing and the like. IPP does not require assembly code to be written, and very small code changes can result in significant changes. The FFT/IFFT operation is realized by utilizing IPP, and the test result shows that the on-line processing is completed by utilizing IPP, so that the performance advantage is remarkable. Preferably, the FFT/IFFT can be accelerated in an IPP manner. These three acceleration modes can be used separately or in combination.

Referring to fig. 2, a schematic diagram of a structure of a 5G terminal general platform optimization system based on GPP provided by the present invention for simulating a large number of terminals, regarding the implementation of simultaneously simulating a large number of terminals on a general platform, a real-time operating system based on Linux low latency version naturally supports a multithreading technology, fig. 2 is a software structure of a general platform for simulating a multi-terminal, where scheduling information to be processed by a simulated terminal every TTI [ TTI is a transmission time interval, each transmission channel (TrCH) corresponds to one service, and since requirements of various services on delay are different, Transmission Time Intervals (TTIs) thereof are different, and TTIs may be 10ms, 20ms, 40ms, or 80ms ], is simplified and modularized (as shown by threads such as UE Thread 0, UE Thread 1, etc.), so as to generate a new Thread for execution, ensure efficiency, and perform real-time response. For some common processes which need to be executed by the 5G System in a coordinated manner, such as System Information (System Information), various Measurement reports (Measurement report), Mobility Management (Mobility Management), and the like, are executed on a new thread in a condition-triggered manner (System Information, Measurement report, Mobility Management, and the like), the design of the whole System separates the real-time processing part from the condition processing part, and ensures that each part can run normally. The System bus is a System bus, a large number of simulation terminal architectures of the general platform of the 5G terminal based on GPP provided by the invention are shown in figure 2, a large number of signaling interactions can be generated in the starting access process of the simulation terminal, if all the simulation terminals operate simultaneously, signaling bursts can be inevitably caused, burst information can not be processed in time, and signaling retransmission can be caused, so that the System is inevitably involved in vicious circle, the operation sequence when the terminals are connected needs to be controlled, the terminals are connected in sequence through the thread communication of all the simulation terminals, one-by-one connection of the simulation terminals is realized, and the purpose of multi-user simultaneous online can be achieved when all the terminals are completely executed.

Based on the method for optimizing the 5G terminal general platform based on the GPP provided by the embodiment, the invention also provides a system for optimizing the 5G terminal general platform based on the GPP. Referring to fig. 3, the GPP-based 5G terminal universal platform optimization system includes:

a priority module 10, configured to assign different priorities to programs of different modules in the base station, and assign the highest priority to a control channel and a related program of a control processing flow; after the task with the high priority is processed, the task with the low priority is processed; specifically, as shown in step S100;

the task segmentation and distribution module 20 is used for segmenting the tasks according to the attribute of each task and distributing the subtasks to different threads; specifically, as described in step S200;

the task execution module 30 is configured to allocate a time budget to each subtask, mark a timestamp for each subtask in the processing flow, and compare the timestamp of each subtask with the allocated time budget to determine whether to continue execution or terminate in advance; as described in step S300.

Further, the GPP-based 5G terminal generic platform supports both static scheduling and dynamic scheduling.

Further, the GPP-based 5G terminal universal platform optimization system further includes:

Further, the priority module 10 is also used for assigning priorities of tasks and threads.

Furthermore, the 5G terminal universal platform optimization system based on GPP adopts an FPGA acceleration unit to build a heterogeneous computing platform, and the FPGA is used for accelerating the baseband signal so as to reduce the computing burden of the universal processor; directly accessing a memory of the general server platform by adopting a DMA (direct memory access) technology through a PCI-E (peripheral component interconnect-express) interface to read and write data so as to realize high-speed data interaction between the general processor and the acceleration unit; and parallel processing of single-instruction multi-path data streams is completed by adopting SIMD (single instruction multiple data) instructions supported by a general processor, wherein the instruction set-based software acceleration method comprises bit-level acceleration, symbol-level acceleration and/or sampling-level acceleration.

Since the specific principle and the detailed technical features of the GPP-based 5G terminal universal platform optimization system are elaborated in the above-mentioned embodiment of the GPP-based 5G terminal universal platform optimization method, no further description is given here.

The division of the functional modules is only used for illustration, and in practical applications, the functions may be distributed by different functional modules according to needs, that is, the functions may be divided into different functional modules to complete all or part of the functions described above.

It will be understood by those skilled in the art that all or part of the processes in the methods of the embodiments described above can be implemented by hardware instructed by a computer (or mobile terminal) program, where the computer (or mobile terminal) program can be stored in a computer (or mobile terminal) readable storage medium, and when executed, the program can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a Random Access Memory (RAM), or the like.

In summary, according to the method and system for optimizing a 5G terminal universal platform based on GPP provided in the present invention, different priorities are assigned to programs of different modules in a base station, and the highest priority is assigned to a control channel and a related program of a control processing flow; after the task with the high priority is processed, the task with the low priority is processed; segmenting the tasks according to the attribute of each task, and distributing the subtasks to different threads; allocating a time budget to each subtask, marking each subtask with a time stamp in the processing flow, and comparing the time stamp of each subtask with the allocated time budget to decide to continue execution or terminate in advance; the method optimizes various problems faced by the open architecture based on the open pure software in software implementation, such as real-time processing of LTE and 5G protocol stacks, HARQ feedback time delay, multi-terminal simulation implementation and the like, has high real-time processing and low HARQ feedback time delay, meets the strict requirements of high real-time performance and low delay in mobile communication, facilitates the implementation of multi-terminal simulation, greatly optimizes the existing general platform, and brings great convenience.

It should be understood that equivalents and modifications of the technical solution and inventive concept thereof may occur to those skilled in the art, and all such modifications and alterations should fall within the scope of the appended claims.

Claims

1. A5G terminal universal platform optimization method based on GPP is characterized by comprising the following steps:

C. allocating a time budget to each subtask, marking each subtask with a time stamp in the processing flow, and comparing the time stamp of each subtask with the allocated time budget to decide to continue execution or terminate in advance;

the step B further comprises the following steps:

b1, allocating the priorities of the tasks and the threads;

b2, preprocessing tasks in background threads;

the step C further comprises the following steps:

recording the processing time of each subtask, and continuing to execute the task when the processing time of the subtask exceeds the time budget; and terminating in advance when the processing time of the subtask does not exceed the time budget.

2. The method of claim 1, wherein the GPP-based 5G terminal generic platform supports both static scheduling and dynamic scheduling.

3. The GPP-based 5G terminal universal platform optimization method according to claim 1, wherein the step C further comprises:

4. The GPP-based 5G terminal general platform optimization method according to claim 1, wherein an FPGA acceleration unit is adopted to build a heterogeneous computing platform, and the FPGA is used to accelerate the baseband signal so as to reduce the computing burden of a general processor; directly accessing a memory of the general server platform by adopting a DMA (direct memory access) technology through a PCI-E (peripheral component interconnect-express) interface to read and write data so as to realize high-speed data interaction between the general processor and the acceleration unit; and parallel processing of single-instruction multi-path data streams is completed by adopting SIMD (single instruction multiple data) instructions supported by a general processor, wherein the instruction set-based software acceleration method comprises bit-level acceleration, symbol-level acceleration and/or sampling-level acceleration.

5. A5G terminal universal platform optimization system based on GPP is characterized by comprising:

the task segmentation and distribution module is used for segmenting the tasks according to the attribute of each task and distributing the subtasks to different threads; the task execution module is used for allocating a time budget to each subtask, marking a time stamp for each subtask in the processing flow, and comparing the time stamp of each subtask with the allocated time budget to decide to continue execution or terminate in advance;

the task segmentation and distribution module is also used for distributing the priorities of the tasks and the threads;

the task segmentation and distribution module also comprises a preprocessing module which is used for preprocessing tasks in background threads;

the task execution module is also used for recording the processing time of each subtask, and when the processing time of the subtask exceeds the time budget, the task is continuously executed; and terminating in advance when the processing time of the subtask does not exceed the time budget.

6. The GPP-based 5G terminal generic platform optimization system of claim 5, further comprising:

7. The GPP-based 5G terminal general platform optimization system according to claim 5, wherein an FPGA acceleration unit is adopted to build a heterogeneous computing platform, and the FPGA is used to accelerate the baseband signal so as to reduce the computing burden of a general processor; directly accessing a memory of the general server platform by adopting a DMA (direct memory access) technology through a PCI-E (peripheral component interconnect-express) interface to read and write data so as to realize high-speed data interaction between the general processor and the acceleration unit; and parallel processing of single-instruction multi-path data streams is completed by adopting SIMD (single instruction multiple data) instructions supported by a general processor, wherein the instruction set-based software acceleration method comprises bit-level acceleration, symbol-level acceleration and/or sampling-level acceleration.